Data driven deep learning method for quantifying groundwater flux in deep fractured aquifers with the fractured rock passive flux meter

Abstract

Movement of groundwater in fractured aquifers is highly variable and depends on many factors besides fracture apertures. Hence, downhole techniques that directly map fracture locations, orientations, apertures, and measure groundwater fluxes are valuable tools. Here, we explored the possibility of using the Fractured Rock Passive Flux Meter (FRPFM) with visible dye component to measure groundwater fluxes and identify geometric fracture parameters through laboratory experiments. For this purpose, we used the deep learning model YOLOv8 to accurately identify the dye marks and to measure their areas $A_{dye}$ and widths $\Delta z_{dye}$ from images of the dyed fabric. Results showed that groundwater fluxes were measured with relative errors of ±23 % and ±16 % based on $\Delta z_{dye}$ and $A_{dye}$, respectively, with an overall relative error of ±20 %. The YOLOv8 model showed very good accuracy by achieving high precision $P$ = 0.99 and recall $R$ = 0.75 for both object detection and mask predictions. The $P$-$R$-curve showed that accuracy can be improved by using more images to train the model.